slide1 n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
Dan Grossman Last Updated: May 2012 PowerPoint Presentation
Download Presentation
Dan Grossman Last Updated: May 2012

Loading in 2 Seconds...

play fullscreen
1 / 28

Dan Grossman Last Updated: May 2012 - PowerPoint PPT Presentation


  • 110 Views
  • Updated on

A Sophomoric Introduction to Shared-Memory Parallelism and Concurrency Lecture 4 Shared-Memory Concurrency & Mutual Exclusion. Dan Grossman Last Updated: May 2012 For more information, see http://www.cs.washington.edu/homes/djg/teachingMaterials/. Toward sharing resources (memory).

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

Dan Grossman Last Updated: May 2012


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
    Presentation Transcript
    1. A Sophomoric Introduction to Shared-Memory Parallelism and ConcurrencyLecture 4Shared-Memory Concurrency & Mutual Exclusion Dan Grossman Last Updated: May 2012 For more information, see http://www.cs.washington.edu/homes/djg/teachingMaterials/

    2. Toward sharing resources (memory) Have been studying parallel algorithms using fork-join • Lower span via parallel tasks Algorithms all had a very simple structure to avoid race conditions • Each thread had memory “only it accessed” • Example: array sub-range • On fork, “loan” some memory to “forkee” and do not access that memory again until after join on the “forkee” Strategy won’t work well when: • Memory accessed by threads is overlapping or unpredictable • Threads are doing independent tasks needing access to same resources (rather than implementing the same algorithm) Sophomoric Parallelism & Concurrency, Lecture 4

    3. Concurrent Programming Concurrency: Correctly and efficiently managing access to shared resources from multiple possibly-simultaneous clients Requires coordination, particularly synchronization to avoid incorrect simultaneous access: make somebody block • join is not what we want • Want to block until another thread is “done using what we need” not “completely done executing” Even correct concurrent applications are usually highly non-deterministic: how threads are scheduled affects what operations from other threads they see when • non-repeatability complicates testing and debugging Sophomoric Parallelism & Concurrency, Lecture 4

    4. Examples Multiple threads: • Processing different bank-account operations • What if 2 threads change the same account at the same time? • Using a shared cache of recent files (e.g., hashtable) • What if 2 threads insert the same file at the same time? • Creating a pipeline (think assembly line) with a queue for handing work to next thread in sequence? • What if enqueuer and dequeuer adjust a circular array queue at the same time? Sophomoric Parallelism & Concurrency, Lecture 4

    5. Why threads? Unlike parallelism, not about implementing algorithms faster But threads still useful for: • Code structure for responsiveness • Example: Respond to GUI events in one thread while another thread is performing an expensive computation • Processor utilization (mask I/O latency) • If 1 thread “goes to disk,” have something else to do • Failure isolation • Convenient structure if want to interleave multiple tasks and do not want an exception in one to stop the other Sophomoric Parallelism & Concurrency, Lecture 4

    6. Sharing, again It is common in concurrent programs that: • Different threads might access the same resources in an unpredictable order or even at about the same time • Program correctness requires that simultaneous access be prevented using synchronization • Simultaneous access is rare • Makes testing difficult • Must be much more disciplined when designing / implementing a concurrent program • Will discuss common idioms known to work Sophomoric Parallelism & Concurrency, Lecture 4

    7. Canonical example classBankAccount { privateintbalance = 0; intgetBalance() { return balance; } void setBalance(intx) { balance = x; } void withdraw(intamount) { intb = getBalance(); if(amount > b) throw newWithdrawTooLargeException(); setBalance(b – amount); } … // other operations like deposit, etc. } Correct code in a single-threaded world Sophomoric Parallelism & Concurrency, Lecture 4

    8. Interleaving Suppose: • Thread T1 calls x.withdraw(100) • Thread T2 calls y.withdraw(100) If second call starts before first finishes, we say the calls interleave • Could happen even with one processor since a thread can be pre-empted at any point for time-slicing If x and y refer to different accounts, no problem • “You cook in your kitchen while I cook in mine” • But if x and y alias, possible trouble… Sophomoric Parallelism & Concurrency, Lecture 4

    9. A bad interleaving Thread 2 Thread 1 intb = getBalance(); if(amount > b) throw new …; setBalance(b – amount); intb = getBalance(); if(amount > b) throw new …; setBalance(b – amount); Time “Lost withdraw” – unhappy bank Interleaved withdraw(100) calls on the same account • Assume initial balance== 150 Sophomoric Parallelism & Concurrency, Lecture 4

    10. Incorrect “fix” void withdraw(intamount) { if(amount > getBalance()) throw newWithdrawTooLargeException(); // maybe balance changed setBalance(getBalance() – amount); } This fixes nothing! • Narrows the problem by one statement • (Not even that since the compiler could turn it back into the old version because you didn’t indicate need to synchronize) • And now a negative balance is possible – why? It is tempting and almost always wrong to fix a bad interleaving by rearranging or repeating operations, such as: Sophomoric Parallelism & Concurrency, Lecture 4

    11. Mutual exclusion Sane fix: Allow at most one thread to withdraw from account A at a time • Exclude other simultaneous operations on A too (e.g., deposit) Called mutual exclusion: One thread using a resource (here: an account) means another thread must wait • a.k.a. critical sections, which technically have other requirements Programmer must implement critical sections • “The compiler” has no idea what interleavings should or should not be allowed in your program • Buy you need language primitives to do it! Sophomoric Parallelism & Concurrency, Lecture 4

    12. Wrong! classBankAccount { privateintbalance = 0; • privatebooleanbusy = false; void withdraw(intamount) { while(busy) { /* “spin-wait” */ } busy = true; intb = getBalance(); if(amount > b) throw newWithdrawTooLargeException(); setBalance(b – amount); busy = false; } // deposit would spin on same boolean } Why can’t we implement our own mutual-exclusion protocol? • It’s technically possible under certain assumptions, but won’t work in real languages anyway Sophomoric Parallelism & Concurrency, Lecture 4

    13. Just moved the problem! Thread 2 Thread 1 • while(busy) { } • busy = true; intb = getBalance(); if(amount > b) throw new …; setBalance(b – amount); • while(busy) { } • busy = true; intb = getBalance(); if(amount > b) throw new …; setBalance(b – amount); Time “Lost withdraw” – unhappy bank Sophomoric Parallelism & Concurrency, Lecture 4

    14. What we need • There are many ways out of this conundrum, but we need help from the language • One basic solution: Locks • Not Java yet, though Java’s approach is similar and slightly more convenient • An ADT with operations: • new: make a new lock, initially “not held” • acquire: blocks if this lock is already currently “held” • Once “not held”, makes lock “held” [all at once!] • release: makes this lock “not held” • If >= 1 threads are blocked on it, exactly 1 will acquire it Sophomoric Parallelism & Concurrency, Lecture 4

    15. Why that works • An ADT with operations new, acquire, release • The lock implementation ensures that given simultaneous acquires and/or releases, a correct thing will happen • Example: Two acquires: one will “win” and one will block • How can this be implemented? • Need to “check if held and if not make held” “all-at-once” • Uses special hardware and O/S support • See computer-architecture or operating-systems course • Here, we take this as a primitive and use it Sophomoric Parallelism & Concurrency, Lecture 4

    16. Almost-correct pseudocode classBankAccount { privateintbalance = 0; • private Lock lk = new Lock(); • … void withdraw(intamount) { lk.acquire(); // may block intb = getBalance(); if(amount > b) throw newWithdrawTooLargeException(); setBalance(b – amount); lk.release(); } // deposit would also acquire/release lk } Sophomoric Parallelism & Concurrency, Lecture 4

    17. Some mistakes if(amount > b) { lk.release(); // hard to remember! throw newWithdrawTooLargeException(); } • A lock is a very primitive mechanism • Still up to you to use correctly to implement critical sections • Incorrect: Use different locks for withdraw and deposit • Mutual exclusion works only when using same lock • balance field is the shared resource being protected • Poor performance: Use same lock for every bank account • No simultaneous operations on different accounts • Incorrect: Forget to release a lock (blocks other threads forever!) • Previous slide is wrong because of the exception possibility! Sophomoric Parallelism & Concurrency, Lecture 4

    18. Other operations • If withdraw and deposit use the same lock, then simultaneous calls to these methods are properly synchronized • But what about getBalance and setBalance? • Assume they are public, which may be reasonable • If they do notacquire the same lock, then a race between setBalance and withdraw could produce a wrong result • If they do acquire the same lock, then withdraw would block forever because it tries to acquire a lock it already has Sophomoric Parallelism & Concurrency, Lecture 4

    19. Re-acquiring locks? intsetBalance1(intx) { • balance = x; • } • intsetBalance2(intx) { • lk.acquire(); • balance = x; • lk.release(); • } • void withdraw(intamount) { lk.acquire(); … setBalance1(b – amount); lk.release(); } • Can’t let outside world call setBalance1 • Can’t have withdraw call setBalance2 • Alternately, we can modify the meaning of the Lock ADT to support re-entrant locks • Java does this • Then just use setBalance2 Sophomoric Parallelism & Concurrency, Lecture 4

    20. Re-entrant lock A re-entrant lock (a.k.a. recursive lock) • “Remembers” • the thread (if any) that currently holds it • a count • When the lock goes from not-held to held, the count is set to 0 • If (code running in) the current holder calls acquire: • it does not block • it increments the count • On release: • if the count is > 0, the count is decremented • if the count is 0, the lock becomes not-held Sophomoric Parallelism & Concurrency, Lecture 4

    21. Re-entrant locks work • intsetBalance(intx) { • lk.acquire(); • balance = x; • lk.release(); • } • void withdraw(intamount) { lk.acquire(); … setBalance1(b – amount); lk.release(); } This simple code works fine provided lkis a reentrant lock Okay to call setBalance directly Okay to call withdraw (won’t block forever) Sophomoric Parallelism & Concurrency, Lecture 4

    22. Now some Java synchronized (expression) { statements } • Evaluatesexpression to an object • Every object (but not primitive types) “is a lock” in Java • Acquires the lock, blocking if necessary • “If you get past the {, you have the lock” • Releases the lock “at the matching }” • Even if control leaves due to throw, return, etc. • Soimpossible to forget to release the lock Java has built-in support for re-entrant locks • Several differences from our pseudocode • Focus on the synchronized statement Sophomoric Parallelism & Concurrency, Lecture 4

    23. Java version #1 (correct but non-idiomatic) classBankAccount { privateintbalance = 0; • privateObjectlk = newObject(); • intgetBalance() • { synchronized (lk) { return balance; } } • void setBalance(intx) • { synchronized (lk) {balance = x; } } void withdraw(intamount) { • synchronized (lk) { • intb = getBalance(); if(amount > b) throw … setBalance(b – amount); } } // deposit would also use synchronized(lk) } Sophomoric Parallelism & Concurrency, Lecture 4

    24. Improving the Java • As written, the lock is private • Might seem like a good idea • But also prevents code in other classes from writing operations that synchronize with the account operations • More idiomatic is to synchronize on this… • Also more convenient: no need to have an extra object Sophomoric Parallelism & Concurrency, Lecture 4

    25. Java version #2 classBankAccount { privateintbalance = 0; • intgetBalance() • { synchronized (this){ return balance; } } • void setBalance(intx) • { synchronized (this){balance = x; } } void withdraw(intamount) { • synchronized (this) { • intb = getBalance(); if(amount > b) throw … setBalance(b – amount); } } // deposit would also use synchronized(this) } Sophomoric Parallelism & Concurrency, Lecture 4

    26. Syntactic sugar Version #2 is slightly poor style because there is a shorter way to say the same thing: Putting synchronized before a method declaration means the entire method body is surrounded by synchronized(this){…} Therefore, version #3 (next slide) means exactly the same thing as version #2 but is more concise Sophomoric Parallelism & Concurrency, Lecture 4

    27. Java version #3 (final version) classBankAccount { privateintbalance = 0; • synchronized intgetBalance() • { return balance; } • synchronized void setBalance(intx) • {balance = x; } • synchronized void withdraw(intamount) { • intb = getBalance(); if(amount > b) throw … setBalance(b – amount); } // deposit would also use synchronized } Sophomoric Parallelism & Concurrency, Lecture 4

    28. More Java notes • Class java.util.concurrent.locks.ReentrantLock works much more like our pseudocode • Often use try { … } finally { … } to avoid forgetting to release the lock if there’s an exception • Also library and/or language support for readers/writer locks and condition variables (future lecture) • Java provides many other features and details. See, for example: • Chapter 14 of CoreJava, Volume 1 by Horstmann/Cornell • Java Concurrency in Practice by Goetz et al Sophomoric Parallelism & Concurrency, Lecture 4